Get Started
Home
Topics
Search
Library
Research questionHow can we distill text-attributed graph data without sacrificing joint text-and-structure performance?Models for text-attributed graphs must learn from node text and graph connectivity together, making full-dataset training costly. A compact distilled set must preserve both sources of signal well enough to support comparable learning.
AI
Machine Learning
Multimodal Models
Natural Language Processing
Latest papersRecent research connected to this question, newest first.TaLK: Text-attributed Graph Dataset Distillation via Coupling Language Model with Graph-Aware KernelThe source concerns text-attributed graph learning with language models and graph neural networks. Evidence comes from multiple text-attributed graph benchmarks, where up to 97% of full-dataset performance is reported using 1% synthetic data; broader tasks and deployment settings are not established.research paper · Sep 2, 2026
Related questions
How can a single graph-learning model handle text-, image-, and multimodal-attributed graphs?How can graph topology inference from node observations be unified with generating new graph instances?How can attribute hypergraphs refine frozen graph-clustering assignments without labels or altering the model or original graph?Can text-to-image models match web-scale performance using smaller, reproducible datasets and models?